#Codex Configuration & Customization

Codex can be customized globally, per project, per profile, and per task.

#Configuration Layers

Configuration is layered from broad to specific:

1. Organization or managed policy
2. User config under CODEX_HOME
3. Project instructions such as AGENTS.md
4. Skills and plugin instructions
5. The current prompt

#Common config.toml Fields

Typical settings include model, reasoning effort, sandbox mode, approval policy, AGENTS discovery, web search behavior, feature flags, and MCP servers.

model = "gpt-5.4"
reasoning_effort = "medium"
sandbox_mode = "workspace-write"
approval_policy = "on-request"

#Temporary Overrides With -c

The CLI can override config values for a single run:

codex -c model=gpt-5.4-mini -c reasoning_effort=low

#Profiles

Profiles let you keep separate presets for different workflows, such as safe local development, read-only planning, or CI automation.

#CODEX_HOME

CODEX_HOME changes where Codex reads config, logs, skills, and state. Use it when you need isolated setups.

#Feature Flags

Feature flags control experimental or staged capabilities. Keep them documented so teams know which behavior is expected.

#Desktop App Settings

The Desktop App exposes many of the same concepts through UI: permissions, model selection, integrations, plugins, and appearance.

#Logs And Diagnostics

Use logs when debugging unexpected behavior, missing instructions, or tool failures. CLI logs and app diagnostics can show which config and project instructions were loaded.

#Next Steps

Read Permissions next.